| import evaluate | |
| from evaluate.utils import infer_gradio_input_types, parse_gradio_data, json_to_string_type, parse_readme | |
| from pathlib import Path | |
| import sys | |
| def launch_gradio_widget(metric): | |
| """Launches metric widget with Gradio.""" | |
| try: | |
| import gradio as gr | |
| except ImportError as error: | |
| raise ImportError("To create a metric widget with Gradio, make sure gradio is installed.") from error | |
| local_path = Path(sys.path[0]) | |
| if isinstance(metric.features, list): | |
| (feature_names, feature_types) = zip(*metric.features[0].items()) | |
| else: | |
| (feature_names, feature_types) = zip(*metric.features.items()) | |
| gradio_input_types = infer_gradio_input_types(feature_types) | |
| def compute(data): | |
| return metric._compute( | |
| model='gpt2', | |
| tasks='wikitext', | |
| **parse_gradio_data(data, gradio_input_types) | |
| ) | |
| iface = gr.Interface( | |
| fn=compute, | |
| inputs=gr.Dataframe( | |
| headers=feature_names, | |
| col_count=len(feature_names), | |
| row_count=1, | |
| datatype=json_to_string_type(gradio_input_types), | |
| ), | |
| outputs=gr.Textbox(label=metric.name), | |
| description=( | |
| metric.info.description + "\nThis metric is computed using the 'gpt2' model on the 'wikitext' task.\n" | |
| "Ensure your input is appropriate for the selected task. " | |
| "If this is a text-based metric, wrap your input in double quotes." | |
| " Alternatively, you can use a JSON-formatted list as input." | |
| ), | |
| title=f"Metric: {metric.name}", | |
| article=parse_readme(local_path / "README.md"), | |
| ) | |
| iface.launch() | |
| module = evaluate.load("d-matrix/dmxMetric") | |
| launch_gradio_widget(module) |